ME3130 QUALITY PLANNING AND ANALYSIS (G Slot)

Statistical Process Control (SPC)

Need Xbar and R charts P chart C chart Applications

SPC
A. Ramesh PhD Department of Mechanical Engineering National Institute of technology Caliccut673601 ramesh@nitc.ac.in

Variation
Variation is natural - it is inherent in the world around us. No two products or service experiences are exactly the same. With a fine enough gauge, all things can be seen to differ. One of the roles of management is work with all employees to reduce variation as much as possible.

The Presence of Variation

8 Measuring Device
Tape Measure Engineer Scale Caliper Elec. Microscope

4
4 4.01 4.009

4
4 4.01 3.987

4
4 4.01 4.012

4
4 4.00 4.004

4.00913

3.98672

4.01204

4.00395

Types of Variation
Common Cause Variation: The variation that naturally occurs and is expected in the system -- normal -- random -- inherent -- stable Special Cause Variation: Variation which is abnormal indicating something out of the ordinary has happened. -- nonrandom -- unstable -- assignable cause variation

Type of Variation Travel Time to Work Example

Measurement of Interest: Time to get to work. Common Cause Variation Sources: -- traffic lights -- traffic patterns -- weather -- departure time Special Cause Variation Sources: -- accidents -- road construction detours -- petrol refills

Statistical Quality Control

Total Product or Process Variation

Total variation = Common Cause + Special Cause To reduce Total Variation First reduce or eliminate special cause variation Reduce common cause variation Identify the source and remove the causes

Measures performance of a process Uses mathematics (i.e., statistics) Involves collecting, organizing, & interpreting data Objective: provide statistical when assignable causes of variation are present Used to
Control the process as products are produced Inspect samples of finished products

Types of Statistical Quality Control

Statistical Quality Control

Quality Characteristics

Variables
Measured values; e.g.,
Acceptance Sampling
weight, length, volume, voltage, current etc.

Process Control

Variables Charts

Attributes Charts

Variables

Attributes

May be in whole or in fractional numbers Continuous random variables

Attributes Has or Has not/Good or Bad/Pass or Fail/Accept or Reject Characteristics for which you focus on defects Categorical or discrete random variables

Statistical Process Control (SPC)

Comparing Distributions
Production Output Example Units Produced

Statistical technique used to ensure process is making product to standard All process are subject to variability
Natural causes: Random variations Assignable causes: Correctable problems Machine wear, unskilled workers, poor material Objective: Identify assignable causes Uses process control charts

Plant A 99 100 100 100 101

X =

Plant B 90 90 100 110 110

500 = 100 5

X
n

500 = 100 5

X =

X
n

No Differences!???

Production Output Distributions

What is the Difference?

Measure of Variation (Sigma) S = Standard Deviation

S=
Plant A X

Frequency

(X X )
n 1
( X X )2

Plant A

Plant B X (X X ) 90 90 -100= -10 90 90 -100= -10 100 100 -100 = 0 110 110 -100 = 10 110 110 -100 = 10
400 = 10 4

(X X )
99-100 = -1 100-100 = 0 100-100 = 0 100-100 = 0 101-100 = 1

( X X )2

90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110

Frequency

Plant B

99 100 100 100 101

=1 02 = 0 02 = 0 02 = 0 12 = 1

12

-102 =100 -102 =100 02 = 0 102 =100 102 =100

= 0
90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110

= 2
S=

= 0 = 400

2 S= = .707 4

The Concept of Stability

99.7% 95% 68%
X 2S = 98.586

Plant A
X + 1S = 100.707

S=

2 = .707 4

X + 3S = 102.121

X - 3S

X - 2S

X - 1S

X
X +1S X +2S X + 3S

X = 100 X 1S = 99.293
X 3S = 97.879
Under Normal Conditions:

X + 2S = 101.414

68 percent of the time output will be between 99.293 and 100.707 units 95 percent of the time output will be between 98.586 and 101.414 units 99.7 percent of the time output will be between 97.879 units and 102.121 units

Plant B
X 2S = 80
X + 1S = 110

S=

400 = 10 4

X + 3S = 130

Control Limits
Control Limits are the statistical boundaries of a process which define the amount of variation that can be considered as normal or inherent variation

X = 100 X 1S = 90
X 3S = 70
Under Normal Conditions:

X + 2S = 120

3 sigma control limits are most common + 3S from the mean If the process is in control, a value outside the control limit will occur only 3 time in 1000 ( 1 - .997 = .003)

68 percent of the time output will be between 90 and 110 units 95 percent of the time output will be between 80 and 120 units 99.7 percent of the time output will be between 70 units and 130 units

Process Control Limits

Special Cause Variation

Relationship Between Population and Sampling Distributions

Three population distributions

Upper Control Limit

Common Cause

UCL= X +3 Average LCL =X - 3

Distribution of sample means

Beta

Normal

Mean of sample means = x x Standard deviation of = x = the sample means n

Uniform
3 x 2 x 1 x x + 1 x + 2 x + 3 x

Lower Control Limit Special Cause Variation

(mean)
95.5% of all x fall within 2 x 99.7% of all x fall within 3 x

Theoretical Basis of Control Charts

Theoretical Basis of Control Charts

Central Limit Theorem

Central Limit Theorem

As sample size gets large enough, sampling distribution becomes almost normal regardless of population distribution.

Mean

Standard deviation

X =

x =

x
n

X =

X
X

Process Control Limit Concepts

Control Limits Define the limits of stability The ULC and LCL are calculated so that, if the process is stable, almost all of the process output will be located within the control limits. 3 sigma control limits The most commonly used UCL is 3 standard deviations above the average LCL is 3 standard deviations below the average If the process is stable, only about 3 out of 1000 process outputs will fall outside the control limits.

Process Control Limit Concepts (continued)

Measures inside control limits are assumed to come from a stable process - Measures outside the control limits are unexpected and considered the result of a special cause The control limits are computed directly from the sample data selected from the process -- The limits and the average are not the choice of management or the operator Formulas exist. The control limits define the range of inherent variation for the process as it currently exists, not how we would like it to be

Control Chart Purposes

Control Chart Types

Show changes in data pattern

e.g., trends
Make corrections before process is out of control

Continuous Numerical Data Variables Charts

Control Charts

Categorical or Discrete Numerical Data Attributes Charts

Show causes of changes in data

Assignable causes
Data outside control limits or trend in data
R Chart

X Chart

P Chart

C Chart

Natural causes
Random variations around average

Statistical Process Control Steps

Chart

Start

Produce Good Provide Service Take Sample

No

Assign. Causes? Yes

Inspect Sample Create Control Chart

Stop Process Find Out Why

Type of variables control chart Shows sample means over time Monitors process average Example: Weigh samples of coffee & compute means of samples; Plot

X Chart Control Limits

UCL x = x + A 2 R LCL x = x A 2 R
Sample Range at Time i From Table

Factors for Computing Control Chart Limits

Sample Size, n 2 3 4 Mean Upper Lower Factor, A2 Range, D4 Range, D3 1.880 3.268 0 1.023 0.729 0.577 0.483 0.419 0.373 0.337 0.308 2.574 2.282 2.115 2.004 1.924 1.864 1.816 1.777 0 0 0 0 0.076 0.136 0.184 0.223
014 . 8

x = i =1 n

xi

Sample Mean at Time i

5 6 7 8 9 10

R =
# Samples

i =1

Ri n

R Chart

R Chart Control Limits

UCL R = D 4 R

Type of variables control chart

Interval or ratio scaled numerical data
LCL R = D 3 R

From Table

Shows sample ranges over time

Difference between smallest & largest values in inspection sample
n

Monitors variability in process Example: Weigh samples of coffee & compute ranges of samples; Plot

R = i =1 n

Ri

Sample Range at Time i # Samples

Out-of-controlwhen? Process is Out of Control

Process Control Chart

200 180 160 140 120 Measure 100 80 60 40 20 0 200

Shift in Process Average

UCL Average LCL

201

202

203

204

205

206

207

208

209

Sam ple Num ber

Process is Out of Control

Trend: 8 or more points moving in the same direction - up or down
Process Control Chart
200 180 160 200 180

Process is Out of Control

Nonrandom Patterns Present in the Data
Process Control Chart

Process Average Trend Up UCL Average LCL

201 202 203 204 205 206 207 208 209 210

160

Measure

140 120 100 80 60 40 20 0 200

Measure

140 120 100 80 60 40 20 0 200 201

UCL Average LCL

202 203 204 205 206 207 208 209 210

Sample Number

Sample Number

Process is Out of Control Signals of Control Problems

Nonrandom Patterns Present in the Data
Process Control Chart
150 140 130

UCL

120 110 100 90 80 70 60 200 201 202 203 204 205 206 207 208 209 210

Average LCL
Sample Number

A point outside the control limits 7 or more points in a row above or below the average (center-line) Shift 8 or more points in a row moving in the same direction, up or down. Trend Nonrandom patterns in the data
Use Common sense and Good Judgment

Measure

Developing X and R Charts Using X and R Process Control Charts

Situation: Boise Cascade is interesting in monitoring the length of logs that arrive at a mill yard. In the long run, they want the average to be 18 feet and the variation should continue to decline The process output measure is length of the logs. An X and R chart will be developed to monitor the log lengths. Define Process Measurement of Interest Determine Subgroup (sample) size (3-6) Determine data gathering methods where, how, who Determine number of subgroups (20-30) Collect Data Compute X and R for each subgroup Plot X and R on separate charts Compute Control Limits Draw Control Limits and Centerline on Charts

Log length Example: Data 30 days (subgroups) -- subgroup size = 4

Day
1
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 26 20 24 17 17 22 24 18 17 20 21 22 16 17

Log Length Data (continued)

Day Log Length (feet)
1
16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 19 19 21 18 20 23 20 18 20 23 22 18 19 21 22

Log Length (feet)

2
18 20 22 16 19 17 12 19 18 23 20 17 17 19 18

3
21 19 18 22 24 15 20 19 20 19 17 21 22 18 15

4
23 19 18 20 17 18 22 17 14 15 21 23 17 19 23

2
17 19 14 17 18 21 20 18 22 22 14 20 20 19 22

3
21 13 17 25 20 23 20 26 23 21 21 18 16 16 19

4
17 16 16 18 19 21 14 15 21 24 19 22 14 20 21

Compute X for Each Subgroup

Compute R for Each Subgroup R = Subgroup High - Subgroup Low

First Subgroup:

X =

X
n

Where: X = the values in the subgroups n = subgroup size

First Subgroup:

16 + 18 + 21 + 23 X1 = = 19.5 4

R1 = 23 - 16 = 7

Log Length Example: Data 30 days (subgroups) -- subgroup size = 4

Day
1
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 26 20 24 17 17 22 24 18 17 20 21 22 16 17

Log Length Data (continued)

Day Log length (feet)
1
16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 19 19 21 18 20 23 20 18 20 23 22 18 19 21 22

Log Length (feet)

2
18 20 22 16 19 17 12 19 18 23 20 17 17 19 18

3
21 19 18 22 24 15 20 19 20 19 17 21 22 18 15

4
23 19 18 20 17 18 22 17 14 15 21 23 17 19 23

Average = X Range = R
19.5 21 19.5 20.5 19.25 16.75 19 19.75 17.5 18.5 19.5 20.5 19.5 18 18.25 7 7 4 8 7 3 10 7 6 8 4 6 5 3 8

2
17 19 14 17 18 21 20 18 22 22 14 20 20 19 22

3
21 13 17 25 20 23 20 26 23 21 21 18 16 16 19

4
1 16 16 18 19 21 14 15 21 24 19 22 14 20 21

Average = X Range = R
18.5 16.75 17 19.5 19.25 22 18.5 19.25 21.5 22.5 19 19.5 17.25 19 21 4 6 7 8 2 2 6 11 3 3 8 4 6 5 3

Plot the X Values

P lot o f S u bgro up Ave ra ge s
50 45 40 35 Subgroup Average

Plot of R Values (Ranges)

Plot of R Values
18 16 14 Range (R) 12 10 8 6 4 2 0 1 3 5 7 9 11 13 15 17 19 21 23 25 27 Subgroup 29

30 25 20 15 10 5 0 11 13 15 17 19 21 23 25 27 Su b g r o u p 29 1 3 5 7 9

Compute Centerlines for Each Chart

Plot the Centerline on X Chart

P lot o f S u bgro up Ave ra ge s
50

X Chart:

X =
R Chart:

Xi
k

Subgroup Average

577.5 = = 19.25 30

45 40 35 30 25 20 15 10 5 0 1 3 5 7 9 11 13 15 17 19 21 23 25 27 Su b g r o u p 29

X = 19.25

R =

R
k

171 = = 5.7 30

Plot of Centerline on R Chart

Plot of R Values
18 16 14 Range (R) 12 10 8 6 4 2 0 1 3 5 7 9 11 13 15 17 19 21 23 25 27 Subgroup 29

Compute the Control Limits on the X Chart

Table n 1 2 3 4 5 6 A2 2.66 1.88 1.02 0.73 0.58 0.48 D3 0.0 0.0 0.0 0.0 0.0 D4 3.27 2.57 2.28 2.11 2.00

R= 5.7

Compute X Control Limits

t Table
n 1 2 3 4 5 6 A2 2.66 1.88 1.02 0.73 0.58 0.48 D3 0.0 0.0 0.0 0.0 0.0 D4 3.27 2.57 2.28 2.11 2.00

Plot Control Limits on X Chart

Plot of Subgroup Avera ge s
30

25

UCL

23.41 X = 19.25 15.09

Subgroup Average

20

15

UCL = X + A 2 R = 19.25 + .73(5.7) = 23.41 LCL = X - A 2 R = 19.25 - .73(5.7) = 15.09

LCL
10

0 1 3 5 7 9 11 13 15 17 19 21 23 25 27 Subgroup 29

Now Plot the Control Limits on the X Chart

Compute Control Limits for R Chart

n 1 2 3 4 5 6 A2 2.66 1.88 1.02 0.73 0.58 0.48 D3 0.0 0.0 0.0 0.0 0.0 D4 3.27 2.57 2.28 2.11 2.00
Range (R)

R Chart with Control Limits

Plot of R Values
18 16 14 12 10 8 6 4 2

UCL

13.0

UCL = D 4 R = 2.28(5.7) = 13.00 LCL = D3 R = 0.00(5.7) = 0.00

Plot the Control Limits on R Chart

5.7 LCL
11 13 15 17 19 21 23 25 27 Subgroup 29 1 3 5 7 9

0.0

P Charts Utilizing the Control Charts

Continue to Collect Subgroup data Plot Values to X and R charts Examine the R Chart First - Then the X Chart Look for Signals A point outside the control limits 7 points in a row above or below the centerline 8 points in a row moving in the same direction any nonrandom patterns Take action when signal indicates Update Control limits when appropriate Used When the Variable of Interest is an Attribute and We are Interested in Monitoring the Proportion of Items in Sample that have this Attribute Can accommodate unequal sample sizes. Sample sizes are usually 50 or greater. Need 20-30 samples to construct the P-chart. Examples: Proportion of Invoices with errors Proportion of Incorrectly Sorted Logs Proportion of Items Requiring Rework

P Chart Example
Plywood veneer is graded when it comes out of the dryer. Sheets that graded incorrectly cause problems later in the process. Management is interested in monitoring the rate of incorrectly graded veneer. The variable of interest is the proportion of incorrectly graded veneer. Each shift, n=100 sheets are selected and evaluated for grade. The number of mis-grades are recorded.

P Charts
Step 1:
Collect appropriate data.
Attribute data of the yes/no type A Sheet is inspected. Is it incorrectly graded - Yes or No? Record the number of Yes occurrences

10

P-Chart Data

P Charts
Step 2:
Calculate the fraction defective for each subgroup.
The fraction defective is known as the p value: number of nonconform ances in the subgroup p= size of the subgroup Key Point:
The fraction defective is always expressed as a decimal value. Using the percentage value (i.e. 4.7% rather than .047) will cause later computations to be inaccurate.

Fraction Nonconformance - p- Values

Step 3:

P Charts

Plot the data on a graph.

Plot each p value

Plot of the p-Values

Step 4:

p Charts

Compute the center line for the p chart and plot on the chart
The center line of the p chart is p

total number of nonconformances in all subgroups p= total number of items examined in all subgroups
p= 429 = .215 2,000

11

P-Values and Centerline

p Charts
Step 5
If the sample sizes are equal, compute the 3 sigma control limits using the following formulas - plot on control chart:

Upper Control Limit

CL = .215

UCL = p + 3 UCL = .215 + 3

p (1 - p ) n = .338

.215(1 .215)

100 Lower Control Limit LCL = p - 3 p (1 - p ) n

.215(1 .215) LCL = .215 3 = .092 100

P Control Chart
UCL = .338

P Charts
Analyzing p Charts
CL=.215

p charts are analyzed using the standard tests for special cause variation:
A Point located outside the control limits 7 or more points above or below the centerline 8 or more points moving in the same direction Other evidence of nonrandom patterns

LCL = .092

P Charts
Step 5: Alternative - When sample sizes are not equal
Compute the 3-sigma upper and lower control limits for the p chart.
If the size of the subgroup size varies, the control limit calculations can be accomplished by two methods: Compute multiple control limits based on the largest and smallest subgroup sizes
The two sets of control limits are plotted on the p chart. By calculating control limits based on the largest and smallest subgroups, both the narrowest limits (largest subgroup size) and the widest limits (smallest subgroup size) are plotted.

p Charts
Using Multiple Control Limits:
In analyzing a control chart with multiple limits, it must be clear that:
Any value plotting outside the widest control limits is considered out of control Any value plotting inside the narrowest control limits is considered in control Only those values, if any, which plot between the two upper or two lower control limits raise questions needing further evaluation (calculate their individual limits)

Compute separate control limits for each fraction nonconformance.

12

c Charts

c Charts
Necessary Characteristics
Subgroups must be the same size (in practical use, if they vary less than + 15% from the average it is acceptable to use the average subgroup size to compute the chart) Subgroup size must be large enough to provide an average of at least 5 nonconformities per subgroup The attribute of interest is the number of nonconformities per unit Each unit may have one or more nonconformities The actual number of nonconformities is small compared with the number of opportunities for nonconformities

A c chart is a process control tool for charting and monitoring the number of attributes per unit. Each unit must be like all other units with respect to size, volume, height, or other measurement.

c Charts
Step 1:
Collect appropriate data.
Attribute data of the counting type Issue is Re-patch requirements. Subgroup size is 3 sheets of plywood Variable of interest is the combined number of repatch spots in the three sheets

C-Chart Example
Boise Cascade Plywood Plant has to re-patch sheets when knot patches become loose or are missed during the initial patch line operation. The department is monitoring the number of re-patches. 3 Sheets are grouped to make sure that the average number > 5

Re-Patch Data
Step 2:

c Charts

Graph the data.

The number of nonconformities is on the vertical axis The sample number is on the horizontal axis

13

Plot The Nonconformities

Step 3:

c Charts
Compute the average number and standard deviation of defects per unit.
Average:

c=

total nonconform ities in all samples number of samples

Standard deviation:

s= c

Compute the Mean and Standard Deviation

c Charts
Step 4:
Compute the 3-sigma upper and lower control limits.
Upper Control Limit

Total = 277

UCL = c + 3 c = 11.08 + 3(3.33) = 21.08

Lower Control Limit

C=

277 = 11.08 25 S = 11.08 = 3.33

LCL = c - 3 c = 11.08 3(3.33) = 1.08

c Charts
Step 5:
Plot the center line, c , and the upper and lower control limits.

c Control Chart
UCL=21.08

CL=11.08

LCL=1.08

14

c Charts
Analyzing c Charts
The c chart utilizes the standard tests for signaling when a process is out of control:
Points located outside the control limits 7 or more points above or below the centerline 8 or more points moving in the same direction Other evidence of nonrandom patterns

c Charts
Common Mistakes
Plotting specification limits instead of control limits Not taking action to determine the special cause when one of the rules for process control has been violated Not plotting the data immediately after it is collected Collecting data on defects per unit when the units are not of the same size, height, etc. Using a desired value to develop control limits rather than actual data from the process

Process Capability
Nominal value Process distribution Lower specification Upper specification

Process Capability
Nominal value Process distribution Lower specification Upper specification

800

1000

1200

Hours

800

1000

1200

Hours

(a) Process is capable

(b) Process is not capable

Process Capability
Nominal value

Process Capability
Light-bulb Production
Upper specification = 1200 hours Lower specification = 800 hours Average life = 900 hours = 48 hours Upper specification - Lower specification

Six sigma Four sigma Two sigma

Lower specification Upper specification

Cp =
Process Capability Ratio

Mean

15

Process Capability
Light-bulb Production
Upper specification = 1200 hours Lower specification = 800 hours Average life = 900 hours = 48 hours

Process Capability Analysis

The process is centered at the target of 200 and the Cp = 2.00. All is well.

Cp =

1200 - 800 6(48)

= 1.39
CP = 1.33 4 Sigma CP =2.0 6 Sigma

Process Capability Ratio

Process Capability Analysis

The process has shifted to an average of 205, but Cp is still at 2.00.

Process Capability Analysis

Real problems exist -- the process is centered at 230. Now, even though Cp = 2.00, much of the output is defective.
Target

Target

Process Capability

Process Capability

Light-bulb Production
Upper specification = 1200 hours Cp = 1.39 Lower specification = 800 hours Average life = 900 hours = 48 hours

Light-bulb Production
Upper specification = 1200 hours Cp = 1.39 Lower specification = 800 hours Average life = 900 hours = 48 hours

Cpk =

Minimum of

x - Lower specification 3

Cpk =

Minimum of

900 - 800 3(48)

= .69 =2.08

Process Capability Index

Upper specification - x 3

Process Capability Index

1200 - 900 3(48)

16

Process Capability

Process Capability

Light-bulb Production
Upper specification = 1200 hours Lower specification = 800 hours Average life = 900 hours = 48 hours

Light-bulb Production
Upper specification = 1200 hours Cp = 1.39 Lower specification = 800 hours Average life = 900 hours = 48 hours

Cpk = 0.69
Process Capability Index

Cp = 1.39
Process Capability Ratio

Cpk =

Minimum of

900 - 800 3(48)

Process Capability Index

1200 - 900 3(48)

17